Share Email Print

Proceedings Paper

New approaches in 3D ultrasound image segmentation
Format Member Price Non-Member Price
PDF $14.40 $18.00

Paper Abstract

Ultrasound images pose some unique challenges to image segmentation algorithms. In this paper we address both the high speckle noise and the adverse attenuation effects of ultrasound. Our results in both synthetic and real images shows significant improvement compared to previous approaches. The approach to both problems is to use Bayesian techniques extended to 3D. The Expectation-Maximization algorithm is used to find the optimal mean and variance of a Gaussian model for the noise, and the segmentation is performed using the Maximization of Posterior Marginals (MPM). Standard application of these algorithms has problems due to the severe attenuation of the ultrasound signal. Our research describes a new solution using a model of the noise with a linear variation in the mean. This is combined with a unique prior probability model to produce improved results.

Paper Details

Date Published:
Proc. SPIE 5016, Computational Imaging, ; doi: 10.1117/12.483901
Show Author Affiliations
Lauren Christopher, Purdue Univ. (United States)
Edward J. Delp, Purdue Univ. (United States)
Charles Bouman, Purdue Univ. (United States)
Charles R. Meyer, Univ. of Michigan (United States)
Paul Carson, Univ. of Michigan (United States)

Published in SPIE Proceedings Vol. 5016:
Computational Imaging
Charles A. Bouman; Robert L. Stevenson, Editor(s)

© SPIE. Terms of Use
Back to Top